AI Made Marketing Faster. Choosing What Matters Got Harder.

Generative AI has removed much of the friction from producing marketing work. Teams can create more headlines, campaign routes, landing pages and audience variants than they could realistically use. That shifts the problem from production to selection: deciding what deserves budget, what should stay consistent and what should never reach the customer.

McKinsey estimates that generative AI could create productivity gains equivalent to 5 to 15 percent of marketing spending. Content production, personalisation and analysis account for much of that potential. Yet Handelsblatt reported that only 11 percent of 200 surveyed retailers and brand manufacturers were already using fully autonomous systems in marketing.

The gap is not difficult to explain because generating material is the easy part. Letting software decide which message represents the company, which customer deserves the budget or which claim should go live carries a different consequence.

A product brief that once produced three campaign routes can now produce 30. The extra 27 cost almost nothing to generate, but someone still has to read them, compare them, check the claims and decide whether any of them answer the original commercial problem. A team that saves two hours writing and spends three hours sorting through variants has not gained much.

The effect becomes clearer in performance marketing. An advertising system can test hundreds of combinations and move spend towards the ones generating the cheapest response. If the instruction is to maximise leads, the system has little reason to distinguish between a customer likely to buy a profitable enterprise contract and someone attracted by an unusually broad offer.

A lower cost per lead can therefore coexist with worse leads. That puts more weight on what happens before the campaign starts. The team has to decide which customer it wants, which conversion matters and what it is willing to pay for it. Once those parameters are clear, AI can optimise inside them. Leave them vague and the system can become very efficient at delivering the wrong result.

More Personalisation Creates Another Problem

AI also removes much of the practical cost of creating different messages for different audiences. A software company can tell a procurement director that its platform reduces risk, tell operations that it saves time and tell management that it lowers cost. Each message may perform well in isolation. Taken far enough, the company stops presenting one proposition and starts presenting whatever each audience is most likely to respond to.

Marketers have always adapted campaigns to different customers. The difference is scale. Producing another 50 versions no longer requires another 50 pieces of manual work.

That makes the fixed elements more important. A company can change the proof point, use case or level of technical detail without changing the reason it wants customers to choose the product. If every part of the proposition becomes available for optimisation, short-term response data can gradually rewrite the positioning.

The problem rarely appears in a single campaign. It becomes visible six months later when the website describes one company, paid media describes another and sales presentations contain claims created for campaigns nobody remembers approving.

Cheap Production Can Make Weak Ideas Last Longer

Previously, production cost killed some ideas before they reached market. A campaign that required a new landing page, video, design work and localisation had to clear a basic threshold before the company spent money producing it.

Teams can now test ideas because testing is cheap rather than because the ideas are good. That sounds harmless until every test requires traffic, budget and attention. Fifty cheap experiments still compete for the same customers.

Before generating variants, the team should know what question the test will answer. Perhaps it wants to establish whether buyers respond more strongly to a productivity claim or a risk argument. Two genuinely different routes can answer that question. Twenty-seven headline variations probably cannot.

Claims Become Easier To Produce Than Evidence

The same imbalance appears in product marketing. Ask a model to strengthen a software proposition and it can quickly turn “helps teams process reports faster” into “cuts reporting time by 40 percent”. The second sentence is better marketing only if somebody can prove it.

As marketing production spreads across AI tools, agencies, sales teams and automated systems, approved evidence needs to become easier to find than invented evidence. A claims register is one practical response. It can record the wording the company has approved, the underlying source, which products or markets it applies to and any conditions that must accompany it.

That sounds administrative until a claim appears in ten places at once. A performance figure introduced in a paid ad can move into a sales deck, website, partner presentation and customer proposal within days. AI accelerates the copying as efficiently as it accelerates the original drafting. The marketing decision happens earlier: whether that number was strong enough to use at all.

The Productivity Gain Depends On What Happens To The Time

Marketing departments will probably produce more with fewer manual production hours. That alone does not establish whether AI has made the function more productive.

A team that reduces campaign production from five days to two could publish twice as much. It could also spend more time interviewing customers, analysing lost sales, improving the offer or removing work that never produced a commercial return.

Those choices produce different businesses as volume remains tempting because it is easy to count. Fifty campaigns, 200 creative variants and ten localised landing pages demonstrate activity. They say little about whether the company reached better customers or gave them a stronger reason to buy. Marketing professionals therefore need to become more demanding about what enters production precisely because production has become easier.

There is no shortage of possible headlines, emails, ads or campaign concepts. Budget, customer attention and the number of messages a market will remember remain limited. Moreover, thecompetitive advantage is not producing everything AI makes possible. It is knowing what deserves to survive the first draft. Make the box much more concrete and tied to the actual problem in the article: too many AI-generated options and not enough discipline about what gets used.

Before You Approve Another AI-Generated Version

Is it solving a different problem?
If two versions make the same point in slightly different words, you probably do not need both.

What are you testing?
Decide whether you are testing the proposition, the audience, the offer or the execution. Do not mix all four.

Which message must stay consistent?
Keep the core proposition fixed unless the purpose of the test is to challenge it.

What result would change the next decision?
Clicks alone may not matter. Use the metric that reflects the customer or commercial outcome you actually want.

Who decides what gets killed?
Set a clear owner for removing weak variants before they reach paid media, the website or the sales team.